Practical question about global frame for 5 independent cameras
Hi community. A have a question and I am looking for comments and/or practical ideas.
Considering 5 independent cameras which FOV are not intersected (so we cannot use epipolar lines). How I can create a global frame (yellow frame at the center of the object)?
How do I know the transformation matrix, rotation and translation, of each camera frame related to a global frame?
Do I need to calibrate the camera and find the intrinsic and extrinsic parameters? If we know the size of the object (gray rectangle) and considering the object has no rotation and the global frame will be set at the middle of it, do I need to use line/circle/corner detection to define the rotation of the camera frames?
Any suggestion or idea will be appreciate. I personally think the answer is easier and just require practical knowledge but I am newbie on computer vision.
You need to have some known size object in each. For example, if you know where the corners are in real-space coordinates you can find the relationship between those four cameras. To find the center camera's location you need something visible in both it and another camera's FOV. For example, a ruler or meter stick. Or if the center of the gray object is marked, you can use that.
I suggest starting with calibrating each camera independently to get the camera matrix and distortion matrix. Then for the four corners, lay a chessboard pattern right at the corner. That gives you a bunch of points at known locations to use with solvePnP. That solves the four corner cameras.
Then very carefully center and align the chessboard pattern at the center and that gives the center camera.
Thank you for your reply. I will start with the calibration using the chessboard.